Droplets size characterization for cold water clouds by means of Generalised Scattering Imaging

This paper describes an experimental investigation carried out at the Italian Aerospace Research Centre (CIRA) laboratory for the application of the Generalized Scattering Imaging (GSI) technique. The droplets were generated by a monodisperse droplet generator and a multi-disperse conical nozzle. This latter is currently used for making the artificial cloud in the CIRA Icing Wind Tunnel (IWT). The measurements are devoted to assessing the application of GSI for the characterization of the cold cloud of CIRA-IWT. The experiments were conducted considering two camera models and demanding technical solutions for the illumination and layout constraints faced for the future implementation of GSI in IWT. To reduce into a linear pattern the enlarged defocused interference fringes, it is found that the optimal aperture of the slit is 5 mm. For the monodisperse droplet distribution, the results show discrepancies between the predicted and the measured diameters regardless of the sharp distribution of the occurrences of the droplet diameters. However, the dispersion found for the droplets generated by the conical nozzle exhibits the expected large variability. This effect is more emphasized when the imaged region is discretized in sub-windows, the number of droplets in each dramatically decreases, however compromising the results of the statistics. Different camera models and imaging configurations led to a variety of measurements at different spatial resolutions.


Introduction
Icing accretion phenomena due to adverse meteorological conditions are considered the main hazards for commercial and general aviation, as elucidated by the regulations emanated by the European Commission through the European Aviation Safety Agency (EASA) [1].The ice accretion due to the encountering cloud condition reported in the certification specification CS-25 appendix C [2] identifies different hazardous conditions related basically to three variables, i.e., the cloud liquid water content, the mean effective diameter of cloud droplets and the ambient air temperature.To release type certifications, EASA verifies the compliance of the products of the manufacturers, i.e., a design organization, concerning the safety regulations under atmospheric icing conditions.To demonstrate the safe operativity of the product, a combined use of numerical simulations, theoretical analyses, ground tests (wind tunnel dry or icing tests) and/or flight tests, is performed.The current design approaches for the air vehicle are mainly based on empirical methods, 2D simulation tools and ongoing experiences gained on in-service aircraft.Even though the current architectures of the aircraft design are responsive to the safety requirements for the occurrence of ice accretion, conservative margins, leading to non-optimized solutions, remain due to uncertainties about the characterization of the ice accretion phenomena.The recent amelioration in the certification regulations introduces Supercooled Large Droplets (SLD), in which water droplets having a diameter larger than 50 microns exist in liquid form at temperatures below 0°C.SLD in cold clouds, i.e., where the temperature is below 0°C, lie in an unstable condition leading to a freezing process when brought into interaction with ice crystals and particles.The hazard concerning the ice accretion over the aircraft surfaces relies upon the occurrence of SLD large enough to hit the exposed surface, having such a mass that prevents the pressure wave traveling ahead from deflecting it.In this event, because of the large size of the SLD, part of it may freeze when in contact with the surface; consequently, the airflow sweeps back the rest of the droplet, triggering the freezing process, as described, until all the SLD completely freezes.This manner of ice accretion forms a transparent, smooth layer, which is difficult to remove, compromising the aerodynamic performance of the aircraft.Hence, radical design solutions are expected for future air vehicle and propulsive system architectures to prevent catastrophic events.To do so, the need of developing validated tools dedicated to the prediction of ice accretion arises to accomplish the industrial demands efficiently and within short development cycles.In this context, the EU-funded ICE-GENESIS project provides the European aeronautical industry with this new generation of three-dimensional (3D) icing engineering tools, i.e., Icing Wind Tunnel (IWT) capabilities and numerical methods.
The capacity to reproduce the icing condition in a controlled environment as an IWT is a crucial aspect of the development and certification of the anti/de-icing systems.Cloud characterization of water content, droplet diameter, and cloud uniformity is of fundamental importance to meet compliance with the requirements emanated by the regulatory authorities.The first two quantities are typically measured by pointwise techniques, e.g., CSIRO King Probe [3] and Phase Doppler Particle Analyzer [4], while the uniformity is addressed by intrusive measurements, using metal grids detecting the ice accretion distribution on a wide surface [5].However, the aforementioned experimental techniques locally elucidate the physical state enabling a pointwise inspection and, intrusively, assessing the level of cloud uniformity.To unveil further insights about the instantaneous spatial distribution of the droplets in the incoming cloud, multi-dimensional sizing techniques are required.
Sizing spherical and transparent droplets through the scattering and imaging of the laser light is referred to in the literature as Interferometric Particle Imaging (IPI) [6], [7].The principle relies on the own feature of the light scattered from the illuminated droplets, i.e., its monochromatism coherence and polarization, being associated with the laser light.A water droplet in the air illuminated by a perpendicularly polarized laser scatters light at two different orders, i.e., reflective and refractive if observed forward at a scattering angle of approximately 65° [6].Two glare points can be identified by imaging the droplet on a camera sensor in focus, and which distance directly relates to the droplet size.However, this direct measurement requires demanding specifications in terms of the camera resolution, which makes currently difficult the implementation.On the other hand, imaging the droplet out of focus, interference fringes form due to the merging of the glare points into a single enlarged spot.The resulting pattern of interference fringes can be related to the droplet diameter, herein described in section 2This method was first proposed by Kӧnig et al. [7], in which the working principle was assessed for pointwise measurements and then supposed to extend the technique to 2D.In their pioneering experiments, the laser light was focused in a region where a train of transparent and mono-disperse droplets was present.A linear array captured the coming light at 10° in forward scattering with respect to the direction of the laser beam.They yielded remarkably accurate results (~2%) and recognized the possibility to apply the technique for more complex droplet distributions such as a spray.Ragucci et al. [8] performed experiments by imaging the interference fringe pattern of a droplet on an intensified camera sensor, oriented on the side at the scattering angle of 90°.Glover et al. [9] extended the technique to perform 2D measurements by illuminating a spray through a laser planar sheet.The droplet light was captured in forward view at a scattering angle of 45°with respect to the laser direction.They explored sprays characterized by droplets with diameters in the range from several up to hundreds of micrometres and number densities ranging from 10 3 up to 10 4 of population.More recently, Maeda et al. [10], [11] presented investigations in which the diameter and velocity of transparent droplets were measured.The novelty lay in a special receiving optical system through which the scattered light passes, inserting in the optical path a cylindrical lens.Hence, this optical compression reduces the circular defocused and enlarged spot, containing the interference fringes, into a linear fringe pattern.This enables the inspection of regions populated by droplets in high concentrations.On the other hand, another technical solution for reducing the circular defocused spot into a linear fringe pattern was adopted by Calabria and Massoli [12], i.e., a linear slit of hundreds of microns of aperture was mounted at the lens, covering most of its aperture.Even though most of the scattered light is cut to the detriment of the signal, it results effectively in its simplicity of implementation.
In this study, the method proposed by Calabria and Massoli [12], namely Generalized Scattering Image (GSI), has been implemented for the preliminary characterization of the water spray nozzle, currently used for generating the artificial cloud in the IWT at the Italian Aerospace Research Centre (CIRA) [13], [14].This study has been performed in the framework of the EU-funded ICE-GENESIS project by placing efforts on the measurements of the droplet diameter and evaluating their spatial uniformity.Herein, the technique has been examined by measuring monodisperse droplets and, subsequently, assessed for the spray.Furthermore, demanding technical solutions, adopted for the future implementation of GSI in IWT, have been addressed.

Theoretical background
As already mentioned, the principle relies on light scattered from the illuminated droplets, i.e., its monochromatism coherence and polarization, being associated with the laser light.In literature, this technique for sizing droplets, each considered spherical in shape, is referred to as Interferometric Particle Imaging (IPI), first proposed by Kӧnig et al. [7].A wise reader can find a detailed and exhaustive description of IPI in [6]; however, only a pertinent argumentation is provided herein.For in-focus images, each droplet exhibits two glare points.An enlarged portion, shown in the picture of Figure 1, illustrates the two intensity peaks for scattered light for each imaged droplet, identifying the glare points.For this optical configuration, these two glare points arise from reflection and first-order refraction.First-order refraction corresponds to the light which has traversed one chord through the droplet.When the imaging optics (e.g., the objective in front of the imaging camera) are placed out of focus the two glare points in the far-field image generate interference fringes, their spacing being proportional to the distance between the glare points on the droplets; hence, the droplet size.The modulation of the interference fringes varies with the absorption of light in the droplet which influences the intensity of the first-order refraction scattering mode.At the scattering angle ϑ = 60° (the scattering angle is the angle between the direction of the laser beam incident on the droplets and the direction of the imaging detection system), the spacing of the fringes, Δϑ, is practically independent by the refractive index of the droplet, i.e., on its composition and temperature [12].Thus, the diameter D of a droplet can be measured by the sizing relation expressed in Eq. (1).
Where αair is the conversion factor which depends on the glare points, the observation angle ϑ, and the relative refractive index between water and air.It is equal to 1.129 for the present experimental setup which set ϑ = 60° [12].Whereas λ is the wavelength of the incident laser beam,  is expressed in radians and has to be extracted from the images.An approach is to evaluate it as formulated in Eq. (2).
Where   is the magnification of the in-focus imaging,   is the pixel pitch, ℎ is the de-focusing distance and   measures the number of fringes per pixel, i.e., the spacing of the interference.This latter is currently evaluated using the Fast Fourier Transform of the intensity signal over the image, using a proprietary software Insight 4G released by TSI®.

Experimental setup
The experimental activities have been conducted at the laboratory of aerodynamic measurement methodologies of CIRA.GSI mainly finds applications in the combustion field, which concerns the characterization of flames generated by burners predominantly working in a laboratory environment.In this context, the main goal is the atomization of the fuel droplets, i.e., of the order of ~10 µm, which are observed at far-field through optical paths relatively short in length.On the other hand, exploring the full envelope of the atmospheric clouds [15] in a harsh environment for icing studies as CIRA-IWT [13], [14], requires investigating droplets in the range between 10 µm to 500 µm.These experimental constraints make the GSI application extremely challenging to implement.To address this purpose, the present study illustrates the experimental setup and the results about the feasibility of GSI to find a possible application for characterizing the water clouds of CIRA-IWT.To do so, the optical distances have been enlarged in reproducing the geometry layout of the test chamber of IWT.
Figure 2 shows the sketch of the experimental setup; from right to left, the laser beam passes through a series of spherical lenses of focal lengths of -50 mm and 100 mm placed at a relative distance such that the focus relies on the observed region.A final cylindrical lens shapes into a light sheet the laser beam.The distance between the optical system and the observed region was set at 1800 mm, this soundly matches half the width of the IWT test section.The GSI system has been composed of an existing PIV system specialized for operating at CIRA-IWT [16].The laser system comprises two Nd-Yag resonator heads equipped with a second harmonic generator providing a laser beam of maximum energy of about 200 mJ at the wavelength of 532 nm and a maximum repetition rate of 10 Hz.The laser heads and the corresponding power units are encapsulated in stainless steel boxes where the environment, i.e., the humidity and the temperature, is regulated.This ensures a stable environment for the devices when they should work in harsh conditions, as reproduced in the IWT test hall.Not reported in Figure 2 for the sake of simplicity, the light sheet is guided in the observed region using a mechanical arm.The system can be commanded remotely, adjusting the laser beam position and the focus of the light sheet.The resulting thickness of the laser sheet is less than 1 mm.
The light scattered from the water droplets is captured by the camera, which is oriented in forward scattering at an angle of 60° and placed at a distance of D1 = 2680 mm from the observed region.This strongly characterizes the present experimental setup resulting in a more representative configuration in terms of geometric constraints encountered in the IWT environment.The observed region is located approximately 120 mm below the orifice/nozzle exit.It enabled to overcome both the overlap between consecutive droplets and the diameter consistency as much as possible.Two cameras were considered for the investigations, i.e., either PCO edge 5.5 or PCO Panda 26 models.Each camera was mounted on a linear traversing system and a motorized rotating stage.A Canon EOS objective of a focal length of 200 mm allows for remote focus adjustment.To increase the magnification of the observed region, a lens extender (EXT), able to increase the focal length by 1.7×, was inserted between the lens optics and the camera.However, this caused a drawback on the total lens aperture f#, increasing that by 1.5 times more than the actual value, e.g., from 2.8 to 4. Based on these considerations, the field of view (FOV) covers different regions of interest, depending on the camera model and optical setup, it ranges between 101.4 × 101.4 mm 2 up to 158.5 × 158.5 mm 2 .This produces a variety of data samples acquired at spatial resolutions ranging from 16.57 pixel/mm to 50.50 pixel/mm.Table 2 summarizes the main parameters of the imaging of the GSI experiments for each imaging considered, where the magnification M is the ratio of the image height over the object height, i.e. in this context, the product between the spatial resolution and the camera pixel size (6.5 µm or 2.5 µm).An optical slit was mounted on the lens, it is a cap that covers most of the entrance.Optimization of the slit aperture has been addressed.To explore the effects inherent in the wideness of the slit aperture, a tailored optical slit was hand-built with simply black paper and knife edges.Once the optimality has a width of the order of 5 mm.This ensures a compromise between the signal amplitude and the fringe contrast.The reference system has the origin at the image center which lays along the nozzle centerlines, the x-axis aligned to the horizontal direction, y-axis along the vertical direction.The z-axis is oriented following the right-hand rule.

Droplet generators
The droplets were issued from either a monodisperse droplet generator or a multi-dispersed conical nozzle, currently used for making the clouds in the CIRA-IWT.The monodisperse droplet generator by TSI, commercially named MDG-100, is a device that exploits the Rayleigh-Plateau instability by applying a constant periodic excitation to a laminar liquid jet inducing surface waves to form and grow.In turn, the breakup into a single droplet per surface wave period occurs.This represents a wellestablished technique for mono-size droplet generation.It is commonly used for fundamental droplet studies like vaporization, combustion, levitation, and surface interaction.The device generates a train of monodisperse distribution of spherical droplets with diameters ranging between 50 µm to 300 µm.The driven parameters are represented by the water flow rate and the excitation frequency.Even though the aperture of the orifice from which the water flow issues plays a key role in the droplet formation, it does not directly contribute to the droplet size.For this application, an orifice with a diameter aperture of 50 µm, a water flow rate of 66 ml/h, and an excitation frequency at 7.43 kHz for having a droplet train featured by a nominal diameter of 167.8 µm.
A multi-disperse spray nozzle was used to investigate a full planar region populated by water droplets characterized by a wide dispersion of the droplet diameters.A standard nozzle belonging to the IWT cloud generator, model SUJ12A055, was considered.

Focus calibration
As already mentioned, one of the features of IPI is the defocusing of the particle image which causes the arising of the interference fringes from the two scattered rays [6].To impose a certain defocus by displacing the recording plane from the focal plane, the camera together with the optical lens can be offplaced concerning the focus location along the optical path.However, considering the constraints imposed by the application of IPI in the context of the IWT facility, the camera and lens were retained fixed at the chosen location.Using a remote control of the focus lens system (EOS controller), the desired defocusing was set at a certain value.Hence, a correspondence between these settings (defined herein, EOS dist.) and the actual focal distance from the target to the camera sensor (CCD-ROI dist. in millimeters) was needed.To do so, a focus calibration was performed by acquiring a set of reference points for each optical configuration.The correspondence between the EOS focus and the focal distance was obtained by fitting a polynomial curve of the 3rd degree of order.Table 3 lists the correspondence between EOS dist.and CCD-ROI dist.for the case of imaging resulting from the 380 mm focal length obtained by the 200mm lens equipped with the extender on the PCO edge 5.5 camera.The calculated distance resulting from the fitting polynomial (fit dist.) as well as the column, diff., indicating the difference (in percentage) between the actual distance (CCD-ROI dist.) and the calculated one (fit dist.) are reported.The maximum mismatch attains 1.42% with respect to the reference at the shortest focus calibration distance.Figure 3 shows the distribution of the focus calibration points in terms of CCD-ROI dist. in millimeters for EOS dist., the fitting curve is superimposed.For the sake of clarity, here the simple calculation of the defocus distance is illustrated.For instance, for glare points imaged at EOS dist.= 960, the actual focal distance corresponds to CCD-ROI dist.= 2691 mm.To obtain the fringe patterns, de-focusing is achieved by setting the EOS controller at EOS dist.= 840, being an actual focal distance reduce down to CCD-ROI dist.= 2432 mm.Hence, the resulting de-focus distance ℎ, i.e., the virtual displacement of the camera and lens system from the optimal in-focus condition, results from the difference between 2691 mm -2432 mm = 258 mm.This parameter is of fundamental importance for the processing of the fringe patterns yielding the corresponding droplet diameters.

Image pre-processing
The enhancement of the image quality has been yielded by pre-processing the raw images captured during the tests.The main goal was the reduction of the background noise level.The level of the background is decreased determining an increment of the visibility of the signal as clearly shown in the comparison plots of Figure 4.As expected, the spacing between the fringes, being 26.4 pixels for both, is not affected by the pre-processing applied to the raw signal as testified by the overlapping shown in the comparison plot.In addition, the enhancement of the quality of the image is also evident by the more visible patterns of the droplets at a low level of intensity.

Accuracy estimation
The evaluation of the uncertainty for the present measurements can be retrieved by following the considerations carried out by Dehaeck and Beeck [17].To do so, recalling the fringe spacing Δϑ found in Eq. ( 2), it can be separated into two different contributions (Eq.( 3). =   •  / (3) Where  / is a conversion factor from pixel to radians.Retaining negligible the uncertainty related to the laser wavelength and applying the propagation of error, the resulting formula is considered for the analysis of the source of errors: Dehaeck and Beeck [17] discussed each source of error in detail.They identified that for the conversion factor αair the sources of error depend on the glare points, the observation angle, the refractive index between the droplets and air, and the non-sphericity of the droplets.On the other hand, the uncertainty related to Fp estimation is not influenced by experimental parameters, only by the level of the signal of the fringe pattern.The uncertainty about Cpix/rad is strictly related to the calibration procedure.In the present context, the calibration procedure considered herein is referred to as two-step theoretical calibration, which procedure is described in detail by Dehaeck and Beeck [17].Thus, the formula for the evaluation of  / reads as: Here, the sources of errors are comprised in the variation of   across the laser sheet and in the measurement of ℎ.A realistic level of uncertainty for the measurements of water droplets attains 1.5% combining all the considered quantities.

2D Spatial discretization
Since the present experiments allow for the estimation of the droplet diameters and their location in a 2D inspected region, it becomes meaningful to evaluate the degree of uniformity of these water droplets, for instance, in the incoming water cloud present in the IWT test section.However, an intrusive approach was presented by Ide and Oldenburg [5], where they assessed the uniformity degree in the quantity of the liquid water within the cloud distributed from wall to wall and from floor to ceiling of the NASA Icing Research Tunnel (IRT).They used a stainless-steel grid installed crosswise in the test section, the ice accreted on the grid during the run.The level of uniformity was inspected by measuring the accretion of the ice along the grid stainless bars.Whereas, in combustion application, Madsen et al. [18] presented measurements of droplet size and velocity distribution using IPI for a swirled water spray.They proposed a discretization of the Region Of Interest (ROI) into different sub-windows, making a spatial structure of the spray.For the present study, similarly, it is possible to virtually discretize in subwindows the ROI.For instance, the imaging obtained using the Panda 26 and the objective lens of 200 mm of focal length covers a final area of 158.5 x 158.5 mm 2 .Each square sub-window measures 16 x 16 mm 2 , where the droplets can be localized and counted in association with their diameters.Once discretized the ROI in sub-windows, the number of occurrences of the droplets lying in each is estimated retaining the correspondence with the measured diameters.This allows for the building-up of a map where the arithmetic average diameter and its standard deviation are retrieved.To do so, a Matlab® script was written for the post processing of the gathered data.

Results and discussion
In the following, the setup of the experiments in exploring the optimized technical solution is presented.The diameter and the level of uniformity of the droplet diameter have been addressed.As already mentioned, the slit aperture in the experiments by Calabria and Massoli [12] was of the order of hundreds of micrometres, which works suitable for experiments where the detection system is placed at a far-field distance relatively short.However, for the present experimental setup, the optimal width of the slit aperture attains to 5 mm.It is worth underlining that the far-field distance is approximately D1 = 2680 mm.This reflects the expected imaging for the application of the GSI technique at the CIRA-IWT facility.In the early imaging, the slit aperture was set in the range of 0.50÷0.80mm; therefore, fringe patterns have been captured with the signal intensity at a low level.Hence, the amount of light coming from the scattered droplets dramatically reduces.Changing the width of the slit aperture, the overall effects on the imaging are represented in the maps of Figure 5; the oscillation patterns are imaged by the light scattered from droplets trains generated by the MDG-100 at three different slit apertures, i.e., 10.8 mm, 5 mm and 2 mm.The map at the left of Figure 5 shows visible oscillation patterns, however, each is characterized by a height more pronounced due to the width of 10.8 mm, being of 16 pixels.At the centre (Figure 5), the slit aperture is reduced down to 5 mm, this causes the reduction of the height of the oscillation patterns resulting in 7 pixels, preserving their intrinsic spacing.A further reduction of the aperture down to 2 mm provokes the diffraction of the incoming light, inducing twofold effects: the height of the patterns increases, and the spacing of the oscillation patterns is compromised.Once the optimal imaging for the GSI was identified, the investigation has been focused on the measurements of the droplet diameter of a monodisperse distribution, as generated by the MDG-100.This makes possible a reliable comparison between the results obtained using the two cameras, PCO edge 5.5 and PCO Panda 26.For each one, the level of laser intensity and the defocus displacement were varied to enhance the fringe contrast.For the PCO edge 5.5 camera equipped with the extender, Figure 6 shows the distribution of the number of occurrences over the measured diameters in which multiple peaks are found.The maximum diameter attains 136.7 µm, below the nominal expected diameter D = 167.8µm.This discrepancy can be ascribed to a bias between the measurements and the value declared by the MDG-100 manufacturer.Furthermore, the multiple peaks over the occurrence distribution indicate instability of the MDG-100 in the generation of the droplets.For the PCO Panda 26 camera equipped with the extender, the number of occurrences along the measured diameters forms a sharp distribution characterized by a single peak at 121.7 µm (Figure 7).Even though the discrepancy between the nominal and measured diameters is still present, the behaviour in generating of monodisperse droplets distribution by the MDG-100 is observed.

[µm]
Tests have been carried out for the conical nozzle SUJ12A055 involving both camera models together with optical configurations, as already given for the distribution of the nominally monodisperse droplets.The inflow condition for the conical nozzle has been set at a constant pressure of air and water, i.e., Pa = 0.2 bar and Pw = 0.4 bar, respectively.Here, the investigation is also focused on the spatial organization of the oscillation patterns imaged in the FOV.To inspect the different levels of density of droplets in the FOV, the investigation has been conducted considering either the core of the spray plume (z = 0 mm), i.e., at the plume centerline, or in the shear region (z = -31 mm off from the centerline).This enabled the inspection at maximum concentration, as shown in Figure 8 (left) and at less droplet density as in Figure 8 (right).It is possible to observe that images characterized by higher droplet density can be inspected despite the overlap between the fringe patterns.As expected, the distribution of the number of occurrences spreads all over the diameter spectrum.For instance, using a PCO edge 5.5 camera the minimum of the standard deviation (std) was reached at 27.2 µm for the test conditions at z = -31mm, i.e., off-centred from the spray centerline.It is worth noting that all the cases in the shear region present low std values; whereas, at the centerline (z = 0) the whole structures of the jet promote a high dispersion.Using the PCO Panda 26, the level of std for each test dramatically increases reaching a minimum of 68.2 µm using the extender on the Canon lens and 92.1 µm without the extender.For these, the corresponding distributions are not shown for the sake of conciseness.As expected, for low levels of laser energy, hence, for low signal levels in the images, multiple peaks establish as shown in the distribution of Figure 9.

Uniformity assessment
The ROI is discretized by 9 × 9 sub-windows as already discussed in section 3.5, the spatial uniformity of the droplet distribution is first inspected for the case of the droplet train generated by the MDG-100.Herein, the measurements are presented for the centreline plane at z = 0 mm.A slightly uniform distribution of the occurrences of droplets along the vertical direction is recognized in Figure 10, where the bi-variate diagram of the number of occurrences over the ROI is depicted.The adopted reference system foresees the measurement plane x-y vertically aligned to the nozzle orifice.The x-axis is horizontally oriented and the y-axis is directed vertically and positive upwards.The origin is located at the image centre.As expected, the maximum of 236 occurrences is reached in the sub-windows located at coordinate x = 0 mm, y = 64 mm), i.e., closer to the nozzle exit.However, for the present test, it is trivial to note the absence of recognized droplets in sub-windows located off from the centreline.As far as the uniformity of the multi-disperse conical nozzle is concerned, the distribution of the number of occurrences for case Pa = 0.2 bar, Pw = 0.4 bar, obtained on 100 images is shown in Figure 12.The measurement plane was aligned with the nozzle exit, at z = 0 mm, corresponding with the highest droplet concentration.Again, the ROI is discretized into 9 × 9 sub-windows, and the occurrences increase all over the Y-axis, where the plume extends.As expected moving along the horizontal direction to its periphery the droplets decrease in number.It is worth noting that the statistics can be based on an ensemble of a maximum of 458 droplets found in the sub-window at coordinates (x = 0 mm, y = -64 mm) and a minimum of 5 droplets at x = 64 mm, y = 64 mm.As shown for the monodisperse droplet distribution, Figure 13 shows the maps of the average (left) and standard deviation (right) of the droplet diameters found in each sub-window.In the inner region -48 mm ≤ x ≤ 48 mm, the average diameter attains to D = 100.5 ± 128.7 µm.

Conclusions
An experimental investigation for measurement of the droplet diameters and locations using GSI has been implemented for the preliminary characterization of the water spray currently used for generating the artificial cloud in CIRA-IWT.Demanding technical solutions have been developed by reproducing the layout constraints faced for the future application of GSI in IWT.Two droplet generators have been considered, i.e., the MDG-100 device has been operated to produce monodisperse droplets and the conical nozzle, model SUJ12A055, to generate multi disperse droplets.A tailored optical slit was manufactured and placed between the camera objective and the investigated region for capturing the fringe patterns.It is found that a slit with aperture width of 5 mm is suitable for the present experimental setup.Two camera models were considered, either PCO edge 5.5 or PCO Panda 26 equipped with different lens systems, allowing several setups for the imaging.In particular, the results have been assessed at different spatial resolutions.The impact on the size of the field of view was consequently considered.A focus calibration procedure has been applied in a more improved version using a 3rd order polynomial gathering more calibration points, reducing the error introduced by the fitting procedure.
A dedicated code written in Matlab® was developed for the pre-processing of acquired images, whereas the measurements of the droplet diameters and their locations were performed using the commercial software Insight 4G® by TSI company.Exploiting the characteristic of the GSI, to measure together with the diameter of the droplets also their spatial position, spatial uniformity has been estimated by discretizing the ROI into sub-windows and classifying the droplets detected for each image into the sub-windows.In this way, the average diameter and the standard deviation have been computed for each sub-region providing a spatial distribution of the measured droplets.
It is found that, considering the droplets generated by the MDG-100, the predicted monodisperse droplets distribution has been yielded for part of the total cases, indicating an unstable behaviour of the generator device for certain operating conditions.Even though a monodisperse distribution is achieved, discrepancies between the predicted and the measured diameters have been observed.
The diameters of the droplets generated by the conical nozzle have been measured for statistical analysis, i.e., the averages and standard deviations together with the distribution peaks.As expected, a large variability of the data is present, even considering high density of droplets.This can be traceable to the limited number of droplets' diameter gathered in each run.This effect is more stressed when the ROI is discretized in sub-windows, the number of droplets in each region dramatically decreases, compromising the results of the statistics.In light of the considerations discussed above where the spatial resolution of each camera configuration is presented, the demanding requirements of having both a spatial resolution sufficient for resolving fringe spacing of droplets diameters up to 500 µm and inspecting a FOV of the order of couples of ten square centimetres, led to the use of the PCO Panda 26 camera model.However, even the use of the PCO edge 5.5 camera 5 allows the GSI measurement at the expense of reduced spatial resolution.The feasibility of the GSI in the geometric condition of CIRA-IWT has been proven.

Figure 1 .
Figure 1.Raw image of three in-focus droplets, each one recognized by two intensity peaks, i.e., glare points.

Figure 3 .
Figure 3. Focus calibration points and fitting curve for the imaging of the configuration at 200 mm + EXT PCO Edge 5.5.

Figure 4 .
Figure 4. Comparison plot of the oscillation pattern of the pre-processed and raw signal.

Figure 5 .
Figure 5.Comparison maps of the oscillation patterns scattered from monodispersed droplet trains at different slit apertures: 10.8 mm, 5 mm and 2 mm on left, middle and right, respectively.

Figure 6 .
Figure 6.GSI results for PCO edge 5.5 + EXT (camera with extender): number of occurrences of the droplet diameters.

Figure 7 .
Figure 7. GSI results for PCO Panda 26 camera with extender: number of occurrences of the droplet diameters.

Figure 8 .
Figure 8. Droplet diameters in micrometers for the cases at the plume centerline (left), or in the shear region (right) for the conical nozzle SUJ12A055; circles indicate the relative measured size the results in yellow are in micrometers.

Figure 9 .
Figure 9. GSI results for PCO edge 5.5 camera with extender: number of occurrences of the droplet diameters measured for the test case at Pa = 0.2 bar, Pw = 0.4 bar.

Figure 10 .
Figure 10.Bivariate diagram of the number of occurrences of the droplets in each sub-window for monodisperse droplets train (PCO Panda 26).The diameters of the droplets, lying in each sub-window, are gathered for computing the ensemble averaging and the standard deviation, the corresponding maps are depicted in Figure11(left) and (right), respectively.Again, the uniformity of the distribution of the average diameter is detected along the vertical direction testifying the concentration of the monodisperse droplets along the vertical train, it attains to D = 135.8±16.2 µm on average.

Figure 11 .
Figure 11.Maps of the average (left) and standard deviation (right) of the diameter of the droplets for the case of monodisperse droplets train (PCO Panda 26).

Figure 12 .
Figure 12.Bivariate diagram of the number of occurrences of the droplets in each sub-window for multi-disperse droplet distribution (Pa = 0.2 bar, Pw = 0.4 bar) (PCO Panda 26).

Figure 13 .
Figure 13.Maps of the average (left) and standard deviation (right) of the droplet diameter for the case of multi-disperse droplets plume (Pa = 0.2 bar, Pw = 0.4 bar).

Table 1 .
Table 1 lists the main characteristics of each camera.Camera main specifications.

Table 2 .
Main ranges of imaging parameters.

Table 3 .
Reference points for the calibration of the imaging of the configuration 200 mm